An Automated System For Classifying Conference Papers

Ngan, Seon Choon Han (2021) An Automated System For Classifying Conference Papers. Final Year Project, UTAR.

Preview

PDF Download (3209Kb) | Preview

Abstract

In the research conference domain, paper assignment process often poses as a timeconsuming and repetitive task for a chairman. A chairman is required to manually review the contents of a research paper, before assigning it to a suitable reviewer. This project is aimed to develop an automated web-based conference paper system for the manual process of assigning papers to reviewers by using classification models. The project is also aimed to select the best classification model for the system, based on an empirical study. The Knowledge Discovery in Databases (KDD) process was followed as a formal data mining methodology where 1000 AI conference papers were carefully collected, pre-processed and transformed to numerical representations through TF-IDF vectorisation. A randomised stratified 5- fold cross validation was then applied on several data mining algorithms and evaluated using the F-measure as a metric. The Support Vector Machine algorithm resulted in the highest F-measure (0.906), followed closely by Logistic Regression (0.903), Random Forest (0.891), Naïve Bayes (0.880), K-Nearest Neighbour (0.831) and lastly, Decision Tree (0.778). Grid search optimisation was then performed but no significant improvements could be observed. The best classification model was then deployed to a web-based research conference system. The web-based system was developed using the Django web framework, based on a system architecture defined in this project called the Enhanced 3-Tier Web-based System with a Data Mining Layer. In conclusion, an automated paper classification system was successfully developed using classification models and its practical usage was demonstrated on a web-based research conference system to help chairmen in assigning papers to suitable reviewers.

Actions (login required)